Method for finding variants from targeted sequencing panels

A computer-implemented method performed by one or more computer processors to identify a low-frequency sequence variant corresponding to a somatic mutation associated with cancer in an enriched nucleic acid sample, the method comprising:

• (a) obtaining (i) a plurality of sequence reads from a sample that has been enriched for a human genomic region associated with cancer, wherein the human genomic region comprises at least a portion of at least one of the following genes:
• PIK3CA, NRAS, KRAS, JAK2, HRAS, FGFR3, FGFR1, EGFR, CDK4, BRAT, RET, PGDFRA, KIT and ERBB2, and (ii) a reference sequence;
• (b) assembling the sequence reads to obtain a plurality of discrete sequence assemblies by:
• (i) constructing a directed weighted graph from overlapping k-mers of the plurality of sequence reads;
• (ii) demarcating the region in each of the plurality of sequence reads where the sequence is deemed reliable;
• (iii) identifying Eulerian paths in through the directed weighted graph by finding a minimal de Bruijin sequence with forbidden strings, wherein the minimal de-Bruijin sequence is determined using the BEST theorem by determining spanning trees of the directed weighted graph;
• and
• (iv) generating, from the identified Euler paths, the plurality of discrete sequence assemblies, wherein each assembly corresponds to a potential variant sequence supported by the sequence reads;
• (c) determining which of the potential variants are true and which are artifacts by:
• (i) evaluating the sequence reads that make up each of the discrete sequence assemblies by examining the sequence reads that make up each of the discrete sequence assemblies;
• (ii) filtering the sequencing reads supporting each discrete assembly by a specified filtering criteria on mapping quality;
• (iii) calculating likelihood scores for each potential variant using a hidden Markov model that evaluates read quality, confidence of base calls, and confidence of alignment of the filtered reads supporting the assembly;
• and,
• (iv) discarding as artifacts those potential variants that are weakly defined based on the likelihood;
• (d) comparing the true variants known to be cancer-associated mutations to the reference sequence;
• and
• (e) outputting a report indicating whether said enriched nucleic acid comprises the low-frequency sequence variant.